Logistic Regression Model Flaw: Accuracy vs Recall in Safety-Critical Industries

Data told me the machine would fail. Physics told me why. I’m continuing my project on building a Predictive Maintenance model for industrial equipment. Last week, I focused on fixing the "Accuracy Paradox." This week, I asked the model, "Which variable is actually causing the trouble?" In Data Science, this is called "Feature Importance." My model flagged "Process Temperature" and "Torque" as the two biggest predictors of failure. This is where the "Mechanical" kicks in. A pure data analyst sees numbers. An engineer sees a story: High Torque suggests the motor is struggling against resistance (it could be a blockage or bearing drag). High Temperature confirms friction or poor heat dissipation. The code points to the symptom. Engineering intuition diagnoses the disease. This is why I believe the future belongs to Domain Experts who code. We shouldn't just predict the crash; we must understand the mechanics behind it. #MechanicalEngineering #DataScience #FeatureImportance #Python #PredictiveMaintenance #Engineering

Moving beyond model.fit()—building Gradient Descent from scratch. 🤖 I’ve been spending time lately digging into the mathematical foundations of Machine Learning. While libraries like Scikit-Learn make it easy to implement linear regression in two lines of code, I wanted to see if I could replicate those results by building a Gradient Descent algorithm from the ground up in Python. In this video, I’m: Defining the cost function (Mean Squared Error). Calculating partial derivatives to update weights ($m$) and bias ($b$). Fine-tuning the learning rate and iterations to reach global minima. Comparing my manual results against the LinearRegression class from Sklearn. The result? A near-perfect match! Understanding the "why" behind the "how" is making me a much better developer as I work on more complex computer vision projects. #MachineLearning #Python #DataScience #GradientDescent #AI #CodingLife

🚀 Understanding Naive Bayes in Action Ever wondered how probabilistic models work? Naive Bayes is a classic generative model that shows the power of reasoning under uncertainty. 🔹 It uses Bayes’ theorem 🔹 Assumes feature independence 🔹 Works surprisingly well even with small datasets 💡Fun fact: It’s often taught using spam classification as an example — not because NB is the cutting-edge choice today, but because it’s perfect for learning core concepts. In my latest Jupyter notebook, I walk through: - Full mathematical derivation - Manual probability calculations with a tiny table - Log probabilities to avoid underflow - Gaussian, Multinomial, and Bernoulli NB variants - Decision boundary visualization - Comparison with Logistic Regression Whether you’re brushing up on ML fundamentals or teaching someone new, NB is a great way to visualize how probability can drive predictions. Check out the full notebook here: [https://lnkd.in/djzpdSCr] #MachineLearning #DataScience #Python #NaiveBayes #LogisticRegression #LinearRegression #HandsOnLearning

Building a Diamond Price Predictor 💎 I just wrapped up a Machine Learning project using Linear Regression to predict diamond prices based on physical attributes. Key steps involved: Data preprocessing & Feature selection. Feature scaling using StandardScaler. Model evaluation with R^2 and MSE. Happy to see a strong correlation in the results! 🚀 💻 GitHub Repository: https://lnkd.in/dzeFdHRn #MachineLearning #Python #DataScience #AI #ScikitLearn

🚀 Learning Project: Logistic Regression from End-to-End (with Visualizations) Recently, I implemented Logistic Regression from scratch using scikit-learn as a learning exercise — not just to train a model, but to understand what’s happening under the hood. 🔍 What I worked on: Used the Iris dataset and converted it into a binary classification problem (Versicolor vs Virginica) Cleaned and encoded categorical labels manually Applied train–test split with stratification Performed feature scaling using StandardScaler Trained a Logistic Regression model Evaluated performance using: Accuracy Precision Recall F1 Score Confusion Matrix 📊 Visualizations I built: Confusion Matrix heatmap Feature importance using logistic regression coefficients Decision boundary (2D projection) Feature distribution histograms Correlation heatmap Model performance comparison chart 🧠 Key takeaways: Feature scaling significantly impacts convergence Logistic regression coefficients give clear interpretability Decision boundaries help visualize how models separate classes Evaluation metrics matter more than accuracy alone This project helped me connect the math, code, and intuition behind logistic regression instead of treating it like a black box. #MachineLearning #LogisticRegression #Python #ScikitLearn #LearningByDoing #MLBasics #Visualization

🚀 Built a Machine Learning Model to Solve a Real Classification Problem Recently worked on an end-to-end ML project where I: • Cleaned and preprocessed raw data • Performed detailed exploratory data analysis • Engineered meaningful features • Trained and evaluated multiple classification models • Optimized performance using proper validation techniques What stood out most? Model performance improved significantly after proper feature engineering and handling class imbalance — not just from switching algorithms. This project reinforced something important: Good ML isn’t about trying every model. It’s about understanding the data first. Tech used: Python, Pandas, Scikit-learn, Matplotlib, SQL More projects coming soon 👀 #MachineLearning #DataAnalytics #Python #AI #LearningInPublic #WomenInTech

🚨 ML Mistake I See All the Time (Even from Pros) You split your dataset. You train your model. Results look great… 🎉 But there’s a silent killer hiding in your code 👉 Class imbalance That’s where stratify comes in. What does stratify mean in Python? In machine learning, stratify ensures that train and test sets keep the same class distribution as the original data. If your dataset is: 70% Class A 30% Class B Both train and test will respect that ratio ✅ The code (simple but powerful): 💥from sklearn.model_selection import train_test_split X_train, X_test, y_train, y_test = train_test_split( X, y, test_size=0.2, stratify=y, random_state=42 )💥 Why it matters: ❌ Without stratify • Missing classes in test data • Fake performance metrics ✅ With stratify • Fair evaluation • Trustworthy results • Better models Rule of thumb: ✔️ Classification → use stratify ❌ Regression → don’t Small parameter. Big impact. Agree? Have you ever been tricked by “good” results that weren’t real? #MachineLearning #Python #DataScience #AI #MLTips #LearningByDoing

Day 7: Leveling up with NumPy! Today’s session at UNLOX® Academy with my mentor Girish Kumar took a deep dive into Numerical Python (NumPy), and it's easy to see why it’s the backbone of data science. We moved beyond basic lists to explore high-performance arrays and data manipulation. What I mastered today: np.arange: Efficiently generating numerical sequences with specific start, stop, and step values. np.reshape: The "magic" of changing data dimensions without altering the data itself. Turning a 1D sequence into a 2D matrix (rows x columns) is a game-changer for organizing datasets. Array Logic: Understanding how multi-dimensional structures power everything from simple tables to complex neural networks. The ability to reshape data instantly makes cleaning and preparing datasets so much faster. Looking forward to putting these tools to work on our next project! 💻📊 #DataScience #NumPy #PythonProgramming #DataAnalytics #UnloxAcademy #TechSkills #BigData #CodingJourney

Day 26 of #100DaysOfAI Continued learning machine learning fundamentals with a focus on probability and statistics concepts used in ML. Today’s focus: Probability basics and random variables Understanding probability distributions more clearly Interpreting p-values and confidence intervals Practicing simple statistical tests and numerical examples in Python What I practiced: Applying statistical concepts to small datasets Interpreting results instead of just calculating them Revising how these ideas are used in model evaluation and A/B testing Keeping the pace steady and focusing on clarity over speed. #Day26 #100DaysOfAI #MachineLearning #Statistics #DataScience

I recently built a Collaborative Filtering–based Recommendation System to predict user movie ratings using the MovieLens ml-latest-small dataset. 🔍 What this project does: Predicts how users would rate unseen movies Learns user preferences & movie features from historical ratings Uses matrix factorization to generate personalized recommendations 🧠 Key Concepts Applied: Collaborative Filtering Cost function with regularization Gradient Descent optimization User–Item interaction matrix 🛠️ Tech Stack: Python NumPy TensorFlow Jupyter Notebook 📊 Dataset: MovieLens ml-latest-small (ratings & movies metadata) 💡 This project helped me understand the math behind recommender systems, not just using libraries but implementing the learning process step-by-step. 📁 Project notebook available on my GitHub (link in comments). #MachineLearning #RecommenderSystems #CollaborativeFiltering #DataScience #AI #MovieLens #Python #LearningByDoing